-
Notifications
You must be signed in to change notification settings - Fork 292
Description
Yes, HeartMuLa is WATERMARKED, as documented in the PAPER, in section No.9, on page 29 ( https://arxiv.org/pdf/2601.10547 )
here is a copy of the text:
9 Ethics and Responsibility
HeartMuLa family is designed as an open-sourced initiative to advance music intelligence research.
Our model operates on a transformative paradigm; it learns statistical acoustic representations to
generate novel musical compositions rather than reproducing copyrighted source material. By
utilizing an exceptionally diverse training dataset, explicitly enriched with culturally and linguistically
diverse music content (see Section 6), our model can innovate and create within various musical
styles effectively, thereby contributing to human musical artistry and cultural heritage.
However, responsible innovation requires more than just advanced architecture. We echo the advocacy
of Ma et al. [2024] for labeling AI-generated/assisted content, thereby ensuring accountability for both
creators and listeners. Notably, we implement a watermarking model to ensure audio security and
facilitate content authentication. We believe this practice is essential for establishing a standardized
responsibility protocol in the rapidly evolving field of AI music generation.
Do not look for the presence of this watermark in the code, since it is probably baked into the weights or incorporated within the VAE, and therefore, invisible to all of us.
We all know that streaming platforms are heading to a ban of the AI-generated music.
Using any part of music created using HeartMula will eventually result in the rejection of any of your work.
SO, What do you think of that?